Robotic navigation system is a well-studied domain, and lane detection is a crucial module for autonomous vehicles. Recently, Unmanned Ground Vehicles (UGVs) have been deployed for several extensive purposes, requiring systems to be better than the traditional heuristic approach to suit specific scenarios. Although lane detecting technology has been developed for decades, many critical challenges in autonomous ground vehicles remain unresolved. This paper addresses the less pervasive but critical challenges posed in a dynamic and complex environment. Thus we chose the Intelligent Ground Vehicle Competition (IGVC) dataset used by an autonomous ground vehicle to navigate through a grass surface with white lanes and obstacles. In the research, the authors developed two core image processing algorithms Open Source Computer Vision (OpenCV), followed by two deep learning models to answer the shortcomings of the OpenCV-based solutions. The deep learning models achieve an accuracy of 98.76% (FusionNet) and 98.68% (Modified UNet) on our test dataset. They are a robust solution to erratic challenges like lighting conditions, a prevalent concern on the grassy surface.
Real-Time Lane Recognition in Dynamic Environment for Intelligent Ground Vehicles
Lect. Notes Electrical Eng.
International Conference on Robotics, Control, Automation and Artificial Intelligence ; 2022 November 24, 2022 - November 26, 2022
2023-11-18
13 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Europäisches Patentamt | 2020
|REAL-TIME LANE CHANGE SELECTION FOR AUTONOMOUS VEHICLES
Europäisches Patentamt | 2022
|